2007/399 inetd backlog SMF property: connection_backlog

[Liane Praza]

Michael Hunter writes:
> The backlog limit pre SMF for unix domain sockets was the same as the
> rest of the services.  For SMF that was changed to 2.  If there was a
> real reason for doing this then you might want to document the seperate
> default.  If not then stick with 10.  The issue is going to be digging
> through whatever old code review or deisgn documentation exists to try
> to understand why the UDS limit was reduced.

Michael and Vineeth and I chatted offline, and believe we have
come to resolution.  A recap for the alias (and Michael can correct
me if I misrepresent)...

The unix domain socket used by inetd is an internal implementation detail
of inetd introduced in S10.  There is no supported way for an
administrator or service author to cause inetd to create a unix domain
socket on their behalf.

The implementors of inetd in S10 (now gone) chose to use a different
 #define for the backlog of that unix domain socket than is used for
the backlog of inetd services.  This is an implementation choice that
may be reasonable, given that inetd owns that socket wholly.  If
it is found to be an unreasonable implementation choice, a bug will
be filed.

In conclusion, this is not an ARC issue, as it pertains wholly to
an internal #define of inetd which doesn't effect inetd services.
Certainly there's no need to document a possible difference between
its value and the backlog used by all other inetd services, as
being introduced in this case.

liane
-- 
Liane Praza, Solaris Kernel Development
liane.praza at sun.com - http://blogs.sun.com/lianep